He's going to talk to us about design.

Or not.

Just what.

You.

All OK so.

Thank you Matthew Can everyone hear
me now yes and this is it working for

the camera man to OK Great OK So yes I
thank you Matthew for the introduction.

So what I mean by micro can
be microbial communities So

basically if you look at nature and you
look at microbes in nature usually they

don't exist in isolation as a sort of
pure species they exist in communities

of multiple species that essentially
interact with each other and

the Senshi the difference pc's
provide benefits to each other and

this creates an incentive for
them to live together in a community

communities that are almost ubiquitous so.

Some examples here are the communities
that are used in wastewater treatment.

The communities.

Of.

Plankton that exist in the oceans we have
our own internal community our digestive

system we're very dependent on
our community for our health and

there are things like microbial mats and
we'll see some of these examples

in more data hours or go through so.

And so as a measured communities are.

Ubiquitous in nature and that clearly
is because being in a community.

Gives advantages so
that suggests that micro

will communities can also be advantageous
in industrial applications and.

If you look at what you market a lot
of the Classical are more traditional.

Applications of biotechnology
they are mainly based on

communities rather than
individual species.

What I really want to talk about
today is the exciting potential for

the design of novel technological
processes based on the systematic design

of micro macro fool communities for
man's purpose.

OK so.

This design problem poses
holds great potential and and

one of this large I'll do as I'll go
through some of the the potential that's

been advocated in a number of papers that
have appeared in the recent years so

you know the primary benefit
is the is the capability to

perform complex multi-step
conversion of chemical products so.

Typically the range of metabolic products
you can achieve from a single market

organism is limited but
obviously if you can.

Use several market organisms that
are that are using the products from

from another organize
them as raw materials and

others that are using the products from
that isn't what ails him in principle you

can create a much more complex synthetic
network and make more complex products.

Obviously a lot of work has been to
been done on the genetic manipulation

of single market organisms
to achieve a goal but

there are limits on that
on what can be done.

And leave the marker awesome stable and
productive.

So.

Consortium so should provide a path maybe
to achieve the same goals that genetic

manipulation can be done without genetic
manipulation or with with only limited

genetic manipulation to each member of the
consortium rather than having to totally

reprogram the genome of a single organism.

A word that a that appeared in
a paper just published this idea

this year is this idea of
what's called emergent bison.

Capabilities So this is the idea
that if you have two marker

on the road working on the substrate they
make you can make certain products but

if you combine those marketing organisms
together working on the same substrate

there's an interaction which
enables them to make a novel and

had a product that wouldn't exist unless
the two organisms were in a consortium and

this opens up the idea of
discovering novel by so

the direct routes that will
be enabled by a consortium so

by being together you can make something
very complex that you couldn't

make if you want your own
on a single substrate.

Another issue which are more more more
serious that will give some discussion

about in this talk of the sort of key
words robustness resilience and stability.

So.

Russell resilience resilience
is perhaps the ability to.

Resist invasion from other species and
your culture our give a good example

of that robustness is obviously
the ability to still produce good outcomes

even though you have a large variations
in the in the inputs of the system and

stability here is often used to
refer to the genome essentially the.

Genome of the species remain stable and

we're capable as a state of
the species remained the same and

you don't have kind of evolution
away from a desired behavior so

after genetic manipulation and robust and

resilience I think particularly are the
key to low cost reductions of biofuels and

chemicals and
go into this issue in much more detail

the final thing I want to
talk about very much is that.

By films are socially
created by consort here and

that's obviously one of the way in which.

Consort here can can grow together
interact and everything and biofilms

provide a way of sort of creating designed
spatial structures for R.T. applications.

OK now.

Looking at this from a sort of design
process engineering perspective.

Communities are inherently dynamic systems

because the world is
constantly changing and

essentially microbes have to respond
to a constantly changing world and.

Resilience the stability of modulating
it through the environment and

the interactions between the organisms but
if we can understand this don't

have any behavior and these interactions
then we can talk about designing or

synthesizing artificial marker but
all colleges for

particular engineering purposes
now perhaps this shows my bias but

as an engineer our I am a great believer
that if if if we're going to take

it undertake a design and
since this task then we have to

really understand the components of
the system that we're going to build

We're going to build a system from and we
have to understand their interactions and

these become the car very complex and
very complicated.

And the ability to sue intuitively
understand what's going on becomes

harder and harder with a system with
more components and more complexity so

we need mathematical models to assist
us arm in understanding a complex

system like this and we're going to use
these to analyze synthesize design and

optimize artificial communities
now a particular thing

that we would like from our mathematical
model is that it's a predictive model so

and I contrast that with say a data is
data very small data base model into.

Because of model that basically just fits
what's already been observed a predictive

model has some cope and some extrapolate
of capability you can go beyond

what's been observed and
can be used to discover novel or

synthesize novel behavior and what I want
to talk about in really the rest of my

talk is a class of models that I
think offer a kind of the ability to

address on all of these
desirable characteristics.

OK so the the model that
the models are going to talk

about are based on this idea
of flux balance analysis

of a marker organism so so
basically what we have is we have a bunch

of candidate species that we want to
consider as candidates to be put under or

consortia and we have to have a basic
model for each species that we're we're

going to be interested in candidates for
our design and flexible and

provides a way of giving these models now

the metabolic network inside of a circle
is a century of very complex reaction

no work then involve thousands of
reactions and thousands of species and

essentially this is the the cells
chemical manufacturing capability this

represents what the cell can
manufacture chemically and.

These days it's very easy to
identify the metabolic network that

exist inside a marker organism at
a at a sort of rough level you

can basically identify it from the genome
so so you can sequence the genome and

mark an organism that enables you to and
identify what enzymes

can be expressed by that genome and
then that tells you which flux is each

basically potential flux in a reaction
network has a specific guidance I'm so

that tells you what basically
connections exist in the network.

And they have also used to build a model
of the synthetic capabilities of the cell

Now obviously there's a lot of
other stuff goes along there are so

you know a lot of the motifs are repeated
in nature so so once you understand

certain basic structures in in
the metabolic network those are repeated

almost every organism in nature and
classes of organisms like say algae will

have a lot of repeated units that
are conservative in evolution so

so what's happened is this it's been
explosion in the availability of these

detailed descriptions of
the metabolisms of Mark organisms and

these this this graph here shows
the socially how of the number and

the size of the model says
in Christian times and

you know a typical state of the are more
will know how involves several files and

fluxes several thousand
reactions in these networks.

OK so.

The flux patterns model sort of
give us a unit operation model for

each individual cell and
arm the it comes in this form of

associate stoichiometry matrix times of
flux vector and what we make is a pseudo

So if you study approximation what we say
is that the metabolism of the cell is in

steady state relative to the dynamics
of its extrasolar environment because

essentially the time scales the solo time
scales are much smaller than the reactor

time scales now just given this
information in from the information about

our constraints on the values
of these fluxes that

particularly these are the fluxes that are
exchanged with the environment there's not

a unique solution to these equations there
are many fluxes that can satisfy those so

the typical hypothesis is to
say is to say that the the cell

has a evolutionary imperative to
maximize its production of by.

Mass so we set up an optimization
problem that the but

the production of biomass is one
of the fluxes in the model and

we say this cell tries to maximize its
biomass subject to its Took a matter

of constraints and
it's strange for us constraints so

this creates an optimization problem that
describes the behavior the cell and it's

a certain special type of optimization
problem called a linear program and

this is about the simplest type of what
an isolation problem you can have so

now from the point of view of a systems
engineer I can start thinking of models of

cells as modules that
basically different pieces and

they're just modules that I can I can
think about and I can work with now

what I want to do is I want to
embed my model of my species into

a say reactor or model some kind of model
of the extracellular environment or

the consortium so a very simple example
might be your world's third reactor and

the battery actor has a set of mass
balance equations that describe.

The how the concentrations of
the various substrates and products.

Change inside the bar reactor and
then them and

then there's a two way interaction between
this model and the model of the species

in the sense that X.
the concentrations in the after

effect the Fluxus exchange
fluxes that are possible and

the solution of the linear program
which is the flux distribution the cell

effects that the mass balances
the dynamic mass balances because

the cell is absorbing substrate and
producing product.

So we we have what we call the resulting
model is what we call a dynamical system

with a linear program invaded we have a
system of ordinary differential equations

are whose right hand sides
are coupled with the.

Solution of a linear program or

a set of linear programs corresponding
to the species in our consortium.

So that's the basic model so.

Let me now talk a bit about one of
the applications we're thinking about

we're developing for for
this type of modeling framework so.

What we're going to look at is the
potential of using algae to make biofuels

and chemicals century to replace petroleum

as the feed stock into
the petrochemical industry.

And algae The big advantage of
ology is is that they in principle

they conceive much greater productivity of
biomass than terrestrial energy costs and

in principle the life cycle can be
carbon neutral because essentially what

the algae is doing is it's taking C O
two out of the out of the atmosphere and

then the sutures being released again
when the car burns the fuel and

the syrup to just recycles
through the atmosphere.

It's not necessary to use freshwater in
fact wastewater of seawater are better

because they're a source of nutrients and
one of the ideas is is

that you could you could you
basically ideal conditions for so

algae farming on on the massive scale
that we would have to do it for

biofuels productions could take
place in coastal desert regions like

Zahar in North Africa where you can
essentially go out into the desert

dig a hole in the sand line it with
plastic pump seawater into it and

grow an algae culture
to make biofuels OK so

now if that sounds that sounds like if
you will idea why are we not doing that

now and the why we're not doing
now is it's currently not

economic to do this it's just that
you cannot compete with the with the.

Current cost per trillion and this is due
to the high cost costs of both the capital

you have to invest and potentially
the substrate that you have to use and

there are two sort of main strategies for
implementing algae farming

one is is what's called
the closed Photo by reactor and

this is essentially a closed you know
highly sophisticated reactor system and

this is a response to the fact that if
I have a algae old all model culture.

It's not able to it hasn't any resilience
to defend against invasion and

competition and predation So
you put it in a sterile.

But that that becomes very
capital intensive to build and

very costly to operate because you
have to keep it clean the algae

will turn to grow on the walls and you
have to sort of scrape it off the walls

otherwise it blocks the sunlight and
reduces effectively so so

proud of our actors are definitely
a feasible technology but

they can be very expensive in terms
of covering the operating costs now

the cheap alternative is
the pond in the desert idea so

this is this is what's called
the pond by reactor and

literally you just it looks a bit
like the stadium in Rome I've

got better pictures later on but
basically you have a paddle wheel and

this pumps water around in
a in a stadium shape and

you introduce an elegy culture here and
then it basically goes round and

grows and then you continuously harvest
algae arm out of the system and

you just have sunlight as shining armor or
on as your energy source Now

the trouble with an open Ponson is that.

Other species and

predators can get in they just get
carried in by the environment and.

The productivity of these systems can be
very very low because basically something

gets in your craw pour out commits your
algae and you don't make any money so

one of the things I want to talk about in
this talk is the idea of crop protection

mechanisms of these cheap open systems and
this is going to be based on the idea

of communities in consortia
to provide crop protection.

Another limitation of open poems
is the carbon limited essentially

the the limiting substrate for most of the
day for the algae is the availability of

C O two in the way the C O two gets in
system it diffuses from the atmosphere and

there's basically not enough here too
in the in the atmosphere to really

stretch the algae the algae could process
much more C O two if it was available so

I'm going to show here
the idea of combining.

An algae with the C O two producing
Narcan organism to increase

the productivity of the production and
there are enough quantities result so but

first let me talk about crop protection so
this is like a long term goal so

this is this is from that appeared
a couple of years ago and

it's a paper by a college asserts
it's quite interesting that

chemical engineers are reading
your columns your papers and

I think it's interesting that
reading the collision five.

So the idea of these guys is that if
you if you look at a natural system so

so Massachusetts in the summer
you can go out into the woods and

you can come across a pond and

sometimes these ponds are really
green stinky fettered palms and

what the green stinky Poland
is is a very productive.

Ector that isn't being
destroyed by predators or

competition so
that's a very resilient ecosystem and

they believe the reason that those
ecosystems are very resilient

is because their consort here it's not
a single species of algae it's actually.

A whole bunch of different species and

a bunch of bacteria species as well
interacting together in a consortia and

it may actually include
certain mild predators in

the overall design of
the of the ecology arm and

the idea is that by having a consortium or
including for

example predators all the ecological
niches in the system are filled so

that if it if another species or
invader comes along arm

it's ecological niches already filled so
it can't get established or

it's much harder for educated stablish so
so a more aggressive predator can't

get established because there's already a
mild predator and system and the idea that

they advocate in this paper is that
you know if you want to build a cheap

algae system then what you should do is
actually design a synthetic consorting or

along these lines that then would give you
the quote protection he would like now.

They talk about this problem
as a qualitative problem.

I'm not quite sure how
they plan to address it

rather than just sort of randomly
putting species together and

trying them out in the lab with our
dynamic flux ballance analysis framework.

It provides us a way to
systematically approach this kind

of design problem the design of
a of a synthetic because gee for

resilient algae growth.

Using modeling and
computation or tools and

hopefully that can save a lot
of experimental effort.

Now over see.

We need.

The scale models to.

To to approach these problems
are we obviously have the and

that at the lowest level
we have our biological and

you can logical model which is
obviously the the flux patterns models

of individual species and the kind
of interactions that can take place.

I mean the species and then there's
an immediate scale we have our sort of bar

reactor which is a sort of chemical
engineering unit operation model but

with these biological
components as part of it and

then at an even bigger level we have how
we're going to design the overall process

that just includes the reactor
as one components.

I would say sort of this part of
the task is pretty well established and

the even a lot of work even on
the algae are in this area so

the key component is being able to sort of
put in the biological part of the problem

in this overall parcel.

And ultimately the idea is that you
know we could be contemplating designing

a console cheer for a specific
engineering purpose we could have sort

of a library of species that we would
have identified that might be useful

we would oversee do all of the literature
search to to identify useful species but

we have a bunch of candidate species
to put it into our consorts here and

then using our modeling and
simulation optimization framework

we can do instead look
at studies to see how.

Different choices for this can sort of
behave we can search over different

different compas different species
to put into the consortia and

we can potentially
systematically optimize.

The conditions on the and the composition
of consortia to achieve our engineering

goal and you know at some point look
that back to experimental validation

arm because models are obviously not
always not always perfect you probably

have to go around this loop several times
before you get a satisfactory solution but

the idea is that going through this loop
is going to involve ultimately a lot less

experimental work rather than just sort of
randomly throwing the species into a B.

current and seeing what happens.

And kind of that kind of just just to
start this kind of a lot of equations but

it's kind of showing.

How you might put together.

A multi species model base
a consortium model based on.

The expanse models for
each individual species OK.

So.

What I'm going to talk about now is is so
so far you've talked about the sort

of big picture vision that we're
we're trying to go after and

now I want to talk about specifically some
of the computational tools that we've

started to develop to address
to address this grand challenge.

So.

Our model in its simplest form is this
idea of a dynamical system with a linear

program so we have differential
equations representing

the behavior of concentrations and so

of of parables in the extra set
of environment and then we have.

A.
Linear program representing all of our

species and
this guy is the is what's called

the solution vector of the linear
program so so kind of there are two

things associated with the sun lotion
of the lunar program one is is the is

the the value of the cost function at
the optimal solution that's a scalar

the solution vector is basically it's
a vector it's that it's the vector of

Fluxus at the optimal
solution of the new program.

And in principle the interaction
is through the solution vector

not necessarily the objective function.

Now.

We came across two key issues
in terms of realizing reliable

technology to simulate models
of this mathematical form arm.

The first issue is that typically in flux
parlance analysis the solution vector

is no not unique it's not that
the singleton set arm and

the the sort of maximizing
biomass argument.

Reduces the non-unique now
split it often doesn't.

Eliminate it or what can happen is you
can start off with a situation where

the solutions are unique and
then you run your simulation and

time passes and suddenly that
the solution becomes non-unique now

the non uniqueness is
a problem because the the.

Differential Equation solves the dynamic
simulators basically want this right hand

side to be unique in new value and if it's
not unique they they basically go haywire.

The second issue is that.

I can start off in a situation
where the the linear program

is feasible which means that you know that
there are there are there are there's at

least one flux vector that
satisfies all constraints but

at some later point in time
it become infeasible and so

it's true that at that point
your simulation has to stop.

And.

Detecting when this happens reliably turns
out to be quite a tricky thing to do

are described in more detail but what
I'm going to show is that an idea called

Hauraki a lot to my station can be used to
address both of these issues and create.

A reliable simulator so.

What's the idea of hierarchical
optimization so forget about the math

just look at the picture so if I have
a linear program the feasible set so

that's a set of feasible solutions
of a lunar program is appalling So

this is that light blue
region that's what apology.

So so any point in this search
is a feasible solution now

when I when I take my objective
function my objective function has this

thing called the cost factor and that's
the gradient of the objective function and

that's a constant factor for a linear
program because there's a linear function

so we can draw the direction of
the cost factor in the space I want to

linear program is trying to
do is it's trying to move.

As far away in the negative direction of
the negative direction of the cost factor

and still remain feasible so we can see
in this simple example if I try and

move in the negative this
direction as far as possible

as they end up here along this line here.

And in this case.

This whole face here of the polyhedral
set is an optimal solution so

this is the case when the optimal solution
is not unique when it's not a singleton so

any of these points minimize is feasible
or minimizes the first object or so

this is the case now what hierarchical
optimization sare this is

I optimized respect to the first
objective and then I pin

my problem of the value of
the optimal first objective and

that essentially reduces the feasible set
for my problem just to this line here and

then I introduce a secondary object if I'm
trying not to my AS respect to that So

in this case my secondary objective
has this point in this direction so

now if I'm forced to sit on this line and
I'm using this object

if that I'm pushed to this point here and
I make the solution unique and

the idea of hierarchical
optimization is if you want to make

the so-called exchange Fluxus which these
are essentially the fluxes that cross

the boundary of the unique so
if I can base what modeling them out.

What I can do is I can set up
the Hauraki your objectives that

if I progressively optimize and then real
optimize and then real optimize I can be

guaranteed at the end of the day
that I make this this vector.

Unique and in that way I can be
guaranteed to have a unique solution for

my problem always now.

That solves the unique but it also raises
a modeling issue because in order to

come up with this model I have to decide
what my Hauraki are for objectives is

you know typically you put biomass
production at the top of the objective.

But what you put a secondary tertiary
quarterly quarterly objectives.

That's a modeling issue and and
really has to be considered

when you're putting together your
model of your Marco Reus and

we've already we've already encountered
situations where you can make good and

bad choices there so you may have
to play around with the model a bit

before you can decide on a good
hierarchy that will give you good

behavior that really reproduces reality.

OK now.

As a matter of the second issue
is infeasible linear programs.

And.

What we're interested
in is is basically how.

The.

The the physical set of the linear

program changes with what is what's
called the right hand side factor so so

all of the influence of the dynamic
parables of the differential equation

on your program can be lumped
into effect to be here and

now we're interested in how this value
function changes is a function of B.

and in fact it's very nicely behaved and
in terms of the way that functions

can behave so it's a convert what's
called a convex piecewise linear.

It's actually Lipschitz continue but
it's only it's defined

essentially on this set airth
which is the set of B. for

which a solution of of the set of
inequalities inequalities exist so the set

of B.'s for which there is a fee that will
satisfy a few pulls be invigorating for

Syria now the problem
with this airth is or

is is that it's what's
called it close set now.

And century or close there is a set

that will contain its boundary
if it has a boundary.

Typical ody theory assumes that
the set on which your right hand so

function is to find is an open set and
an open set as it is a set that doesn't

contain its boundary if it has a boundary.

So and
we'll see why this this close set so

she creates creates us a challenge for
for for applying standard

numerical methods that essentially assume
that there's an open set there and

essentially the linear program is
only going to be feasible while my

my simulation arm makes the right
hand side stay inside that set.

This is easier explained of
the picture OK So so this this F.

is my set of right hand sides that.

Create a feasible LP and.

And basically it's a set that contains its
boundary that's what the blue boundary

line shows now as an example this is this
is this this is a simple simulation of.

I think it's.

I for mounting a glucose silo
some extra arm and what I'm

showing here is the is the concentration
of the very the two substrates of the by.

In the culture and what happens
is that it's got a preferably.

The glucose first so
it doesn't touch the silos and

it starts eating the glucose so that sort
of might correspond to it starting here

in the F circuit and
it carries along happily eating

glucose I'm happy to have you happily
until it reached this point here and

this point here is when it's used up
all it's not glucose at that point

what it does is it switches its
metabolism and it starts eating Xylo So

that's when it starts eating Siler
now now the issue is that kind of in

comparison to eating glucose eatings is a
bit marginal for the cell that's a bit and

so when you read things that are low so
you actually sit on the boundary of this

is feasible said and
you sort of cruise along the boundary.

Happily until this point here which
is when you've eaten up all of us and

at that point kind of the cells die
because it's got no more substrate have to

go or
really what happens is it can't maintain

a steady state metabolism because
there's nothing left to eat anymore but

you know that in the in this list
here that's called cell death so so

this is the point where the simulation
really becomes unfeasible

this is where the part where you
should really stop the simulation.

The trouble is with.

Many of the naive approaches that have
been pro proposed to simulate these

systems the simulation often fails around
here where you bump into the boundary and

the reason the simulation fails is
the numerical methods again and

this is based on this open set idea
many of the numerical methods in

order to do what they need to do they're
sort of sampling around this point and

this sampling process kind of
means that you could sample.

Infeasible point outside the set at
which point your linear program says

I mean feasible and then the numerical the
numerical integrated doesn't know how to

deal with that information and there are
sort of hacks to try to make this work but

typically what after what
you see is that fails or

it sort of it some methods
can sort of crawl along but

it takes an enormous amount of effort and
how.

What we've come up with this is kind of.

A sort of more elegant solution of the
problem that creates a much more reliable

numerical solution and
what we do is essentially we

extend the demand on which
the linear program is defined and

we extend it to the whole space so that's
what this picture is trying to solve so

an idea in Linear Programming is this
thing called the Phase one linear program.

And the idea the role of the Phase
one linear program has in algorithms

is to find a feasible point if one
exists and the way you do it is you set

up a problem that's always guaranteed
to be feasible but if you can find

a certain solution of that problem then it
will be a feasible point and that's Dan

by introducing these new variables air
switcher called slack variables and

you can see that S S equals B. and

vehicle zero is always a feasible
solution of these set of constraints so

this this problem always
has a feasible solution.

Now to find a feasible point what you
truck what you do is you minimize

the sum of the slacks and
the slats are non-negative so

that the smallest you can make that
is zero but if you can make the slack

zero you found of the that satisfies
the constraints on the fees so

the fray essentially phase one LP
is another LP that you solve it.

It's guaranteed to be feasible and if the
optimal solution comes out as zero then

you found a feasible point if it's greater
than zero you know it's unfeasible.

And the way we solve the feasibility
issue is that we put the Phase one

objective in as the sort of zeroth
level now the first level and

so before we consider biomass we
conserve the feasibility problem and

if we have the feasibility problem at the
top of the hierarchy the probe the program

is feasible everywhere I can I can
basically simulate I can simulate I can

simulate this whole trajectory and
I can even define a sort of an extension

of after it becomes the actual problem
comes in feasible and that makes

the simulation very very reliably because
this is no longer a boundary of the set so

it's fine if I sample
around these points and

what also means is that is in
the simulator you can basically monitor

the value of the sky and while the sky
zero you know it's feasible in the moment

it comes off zero you know you've become
an infeasible So it's very easy to see

when you hit the death the death face and
the simulation.

OK so.

That's the boring bit of the talk now
we'll do some simulations now we've got

this this simulation capability
you will use it to simulate it so

so this is a better picture
of one of these raceway.

And you say it's like it's like
a stadium in Rome and you basically.

You know introduce a culture and here
you've got this paddle wheel it pushes

it pushes the water around and then you
harvest again just before it's come around

again and it's essentially continuously
faired and continuously harvested.

You pull off you pull off so

initially when you harvest you
pull off the cells and you.

Brake the the cells are up and down if
you've operative correctly the algae

have sort of accumulated
Leopard's inside them and

that's the leopard's of a major
product if you're making a biofuel but

you also can recycle some
of the substrate so so

first approximation you can think
of this as a plug for a reactor and

you can sort of just approximate
it as a sequence of continuous

terse with a recycle and then we have
sunlight that's driving the system so

if we look at sort of one of our C.S.T.
are in the church chain here.

Are can sort your model.

Is going to be a model that
metabolic reconstruction

of the algae and then the other organisms
I want to put on the heart and sort here.

Then there will be exchange of of farm

substrates between the members
of the console tier and

that's going to be inside a well
stirred reactor model and then

the mass balance of the well star reactor
will involve inlet and outlet terms.

Sirrah to an oxygen exchange for
their atmosphere and a sort of model for

sunlight and pinching on the reactor and
the key thing to note here

is that the sunlight is periodic or
twenty four hour cycle and

we have light and dark faces so this is
this is a model of light intensity but

during the course of the day.

OK so.

The model we're going to use for algae is
is this thing this is a it's a detailed

model this is a large scale metabolic
model of something called climbing on us

and heart I all I can say is it's the
fruit fly of algae it's that it's the one

that everybody started and that's why
there's so much information about it and

and what this is a very complex model but
and it clears a lot of features but

most importantly basically when
there's light it's going to use to.

Solve the C O two and
acetate is carbon source and

it's going to produce oxygen
during the dark period

it consumes acetate to maintain
our maintain it's metabolism and

it produces zero to an Associates which
is between these different metabolic.

Modes.

OK now.

Because this system is forced by
sunlight it doesn't have a steady state

it has something called
a slightly steady state so

what that means is that if you run it for
long enough and

you look at how the concentrations and
everything else changes in the.

Over twenty four hour period
the wave becomes identical you just

repeat the same basically the initial
condition at the beginning of the day

equals the final condition of the end of
the day you just repeat that cycle so

I'm showing you some of those why I've
waveforms at the site itself and how

in this case the way you get the steady
state is you start it from some initial

position and then you just run it long
enough till it settles down to the sucker

state there are more sophisticated ways of
doing that but that's what we did here.

But these are relatively quick simulations
to do it doesn't take that long to do.

So this is the result of doing
a monoculture with algae and

I guess the kind of important things
to see is that that this is sort of

during the night the population
the the concentration of algae goes

down because of the outflow out of
the system and the algae isn't growing but

then when the sunlight comes back
the algae starts growing again and

you have this sort of this is
this is roughly dollar and

this is roughly dusk OK and.

And.

Another interesting feature is how
the concentration of dissolved oxygen and

dissolved C O two change in the bar to.

Of the day the most
important thing is that

you know what happens during that
during most of the day is the C O two

concentration is driven to a very very low
value with the that the algae is limited

by the social mass transfer of
Syria to from the atmosphere.

So.

So the idea here is to create a consort
year by introducing another organism that

produces C O two in situ and that can
boost the growth of the algae arm and

what I'm going to do here is going because
it was convenient to model is good because

there are good models I'm going to use
a model of yeast and again there's a very

well developed model of
yeast which is large scale.

And this will exhibit both aerobic and
anaerobic growth and so

she requires glucose is the carbon source
so so the feed the yeast is going to eat

glucose and produce C O two which
then the algae is going to eat.

So this is what happens in the coal
culture so now we've got both yeast and

algae present and
the key thing to spot here is that

the the the algae concentration
is increased by five more

than fifty percent so and we didn't do
it we haven't done the optimization here

this was kind of just the first simulation
we run so so essentially the introduction

of the yeast boosts the for
productivity of the use substantially and

what you see in terms of
the dissolved gases is that.

The Especially there's
more C O two available so

during the day the increased
availability of Syria to really boosts

the productivity of the of the of
the algae and more generous

also produced as well.

OK.

So aren't going to skip over that so.

So another thing that's got another
thing that's going on here.

Maybe I should explain that so so

what what I'm showing here is two
different simulations in which I've

changed my hierarchy of objectives
I'm outta my station and

basically what's happened with the dotted
line is I made a bad choice and

you see suddenly that in simulation too
there's virtually no oxygen present a top.

And what happened here is that we.

What we said is that one of
the old stories I think that

the one of the objects is for the algae
is to maximize oxygen consumption

whereas this line is minimizing oxygen
consumption so the algae both consumes and

produces oxygen and
if it maximizes oxygen combustion

there's this limitation the model
in the sense that the the ton

the the flux of hydrogen
ions into the algae

is not limited so
if it can absolve if it can.

Suck in unlimited amounts of oxygen
it just makes water to make energy so

that's really a limitation of the model so

what so so we had to go to the we
had to go the other objective but

we also had to introduce a model for
the PH of the solution arm so

that we were we were getting the hydrogen
ion concentration correct so

that involved introducing a bunch of
equilibrium equations corresponding

to the ON IT chemistry taking
place in the in the air and

that that makes the model
a so-called differential algebraic

equation system with an linear program
value because now we have a nobody we have

the algebra chips and
we have the linear programs but given

the tool we developed it was very easy to
take what we have developed an interface.

That two of the DA solve
a problem nobody saw of and

this was all done with a tool
that we developed to Matlab and

it really was a matter of
of an afternoon's work for

my questions to make this change in the
model because the that the competition at

all was so flexible and
so when you go in and

you put the PH buttons into the model
you don't see a dramatic change but

you can argue that that's more realistic.

Are.

And but now you get an idea of how
the the concentrations of the various ions

are changing also in the PH and
the culture.

So this shows now before I just show
the results for one serious three

are now I'm pushing for serious start
serious three hours in a chain so

that's obviously a bigger simulation I got
two species in foresee astri arse and you

can see as as you go along the trail of
C S three rs your concentration increases

and that again you get the dissolved
gases in each of the four seriously ass.

OK.

I was going to tell you a bit about
the the the microbial mat and

the other Stone National Park survey
sickly it's a model of this top layer

of that this is this is basically a source
large that grows on the bottom of

the of the thermal ponds and and and

there's a there's a consort here
of three species in that top lair

focus of this only takes place in that
top layer because like carbon penetrate.

And so these are you know I'll
skip through this is a it's a.

Lot a lot of simulation results here but
the idea is that the if

you start this this model from a bunch
of different initial conditions or

you just run it I think if
you see what we're doing is.

We're running it for mostly here on
the day night cycle what you find

is that is that it always approaches
the same site city state and

what's very interesting about this
model is it predicts the experimentally

observed ratio of two of
the species in the consortia and

previous models of always we've always
had to provide that ratio as an input or

is this model predicts that parameter.

OK so.

I.
Don't really have time to talk about

sensitivity No it's not my station but for
those of you who care about those things

we're working on on our own techniques
to to use these simulations and

more complex situations but
I'll go to my conclusions.

So.

So what I've talked about is
some of the motivation and

some of the sort of technology
of doing details modeling and

simulation of by a process is based
on microbial consort here and

the clear idea behind this is
dynamic facts plans analysis and

we've made numerical
importations of our of our

numerical tools available from
the web page of much of my group and

there's a tool in Fortran which
is more difficult to use but

it's much more efficient and
could do much bigger problems but

there's also a tool to Matlab
which is much easier to use.

And I mean if you want to get into this
side of care if you use the matlab at

all and there and then if you get hooked
on them then you can become a big boy and

work on Fortran and or girl.

And so on and I didn't really have
time to talk about some of the that

the more sophisticated develops in terms
of sensitivity analysis not my station

arm I think there's a there's
a lot of potential for

future work in this area are we
there's a lot of work in.

So she doing systematic mathematical
programming based optimization

of these type of models
on the big issue is that.

The models are are non differentiable So
we have to do with one smoothness.

Are if we if we have a systematic
optimization capability we can use

those to optimize the operation of power
chemical processes based on single or

multiple species we can use it to design
artificial or cultures where we try and

combine different combinations
species together and

we can also use this to actually
identify the moles in the first place so

it can be used to design optimal
experiments for model scrutinization and

parameter estimation for actually building
the metabolic models in the first place.

If you think about the extra
layer of Vironment

the extracellular environment can be
described by a range of different

types of differential equations
they can be O.D.S. different.

Partial Differential Equations
partial differential algebra and so

they're all of these all of these kind of
numerical models that we need to deal with

can be extended to include Lps and.

This can create some very very
large scale simulation problems so

to give an example of this I'm starting
a collaboration with a colleague at MIT

who is an oceanographer and
he's interested in how our own

populations the populations of algae
of different species in the ocean

change with say death in the ocean and
apparently there are phenomena where

you know algae sort of float up and down
in the depths and that and the different

species go to different depths and this
happens on a day night cycle so that they

can explore various changes taking place
and and you if you have a P.D.F. model of

the of the flow of the ocean on the heat
transfer on the ocean you could put in.

Algae models and
models of other of other kind of.

Punkin species and
you can start to understand Tom

some of these phenomena
are in a lot of detail.

I think another interesting question
is that there are more sophisticated.

For experiment small noses that involve
other types of convex hull to my social

problems and we need to develop the theory
and algorithms for this and of course.

The applications of this
seem to be you know

enormous we talk about Barfield
production we talk about.

Bacterial mats we talk about oceanography.

You can talk about bio fuel
productions from sugars etc etc etc.

OK so.

I'll just acknowledge my group members.

As a post-doc he's taken
the leadership role on this project or

microchip Jose income ill who are Ph D.
students of contributed various

aspects of the project and also like to
make a big thank you to my cancer not your

mass who got me interested in
this topic in the first place so

thank you for your attention and.

Thanks.

But.

I think.

You know.

That.

Yes OK So so there's kind of a broader
chemical in engendering question right so

so the in principle when you take the.

Out you harvest the lip hurts and
then what's left is carbohydrates and

proteins and some other stuff.

In principle those carbohydrates could
then be processed to create chokers

doesn't have to be a group for something
that something that the the the other

organism pretty good but if you look at
the overall carbon balance the overall

carbon balance is limited by the rate that
C O two going to fuse into the pond so

so the question the quicker
the question is you know.

If you want to have a completely closed
system that doesn't have an import of

a chauffeur from outside you're probably
limited of how much sugar you can use but

you could also you know you know we
haven't done that calculation on

the overall carbon balance so

I don't know where you know what
that whether that's limiting or not.

But again you know I mean
glucose is a very expensive

carbon source you could think about
importing something much cheaper and

choosing your organism in such
a way to process that so so

I mean ultimately the you know if you
want to go beyond the carbon balance

the economics would decide you know
whether it's worth introducing more

substrate.

Yeah I think those are all
great questions it's.

You know I think that you know it's.

I think as time goes on
we'll start having you know

ways to address those questions
systematically You know I think the.

You know I think that the the idea
of ecological niches is important.

As.
So that gives you some guidance you know

what the ecological niches are and
try to make sure that they're filled.

There are also things like
nutrient requirements.

Either you know do you
want to supply nitrogen or

do you want to introduce an organism
that can produce nitrogen.

Which it will be twelve is a very
important thing as well so

basically there are only certain bacteria
that can make the trial so you know if you

start worrying about B twelve you're going
to have to have something that produces B.

twelve and you're in the open sort of so
so you know.

You know it's a it's almost like
a it's a classic chemical engineering

design problem right you you you have
to think about what you need and

what ecological niches you need to fill
the new you have to start thinking about

what species you want to do that and

what are the potential
interactions what's I mean.

For those of you know what superstructure
optimization that might be a way to

sort of set up a big growth there.

But you know that there are I think there
are potential ways to set that up as a as

a kind of interrupt my social problem
to sort of at least generate kind

of.

How the model share them yeah.

Well I mean there are I mean there
are a couple of ways that can happen.

The picture helps right for.

The typical way is you
know a socially all of

these are metabolites and
they're basically excrete it by one.

Species into their extrasolar medium and

then they were absorbed out of their
exercise or medium by another species so

so the the interaction is essentially
through the reactor dynamics.

But there are other.

Interactions sometimes that are more.

Controlled So basically species
will sort of be physically

located you know they they may actually
kind of like what you know there are algae

that will hot basically shelter.

Bacteria on their cell walls because you
know an algae is enormous in comparison

to a bacteria so this is happens I
think with the B. twelve producing.

Bacteria and they're the interaction
is much kind of more

direct it's not through
the interstellar medium.

But you can you basically model
that with a knowledge of what's

called not a broken strain rather than the
than a dynamic mass balance constraint but

you can model either type of interaction.

Yes Yeah well yes you know because kind
of here are right on the if you look

at these two guys right
these guys are kind of.

You know taking both acetate and ammonia.

That are produced by either the.

Earth side of bacteria so
they they will compete they compete for

the substrate now and and
you know that kind of like the so

the saga study that state you see depends
on how that that competition plays out

so you can you know you can you can
produce different steady state populations

depending on the conditions of the system
because that competition plays out.